Biologically Inspired Computing

Development of Bio-inspired Cortical Feature Maps for
Robot Sensorimotor Controllers 
(2009 - 2013)

An important goal in robotics is to create autonomous machines that can perform basic tasks related to their own maintenance and welfare without human intervention. In short, it would be desirable for them to have more sophisticated human-like capabilities. This raises some major challenges as traditional computing and engineering approaches can only achieve so much. They can be used to mimic human capabilities to some extent but it is difficult to make systems that work in the same way as natural ones do. As we do not fully understand all the neural processing which generates our own behaviour it is often difficult to translate the concepts into traditional approaches.

The main aim of this research project was the transfer of novel principles from the field of Computational Neuroscience to a robotics scenario  where a small, autonomous humanoid robot learns basic visuomotor coordination skills via an approach inspired by how the real mammalian cortex develops. This involved the implementation of a multi-stage developmental learning process inspired by human brain development which included activity independent, activity dependent and lifelong learning phases.

Natural neural systems manage to achieve speed, fault tolerance and flexibility despite having very low power requirements. Therefore, I take the view that it seems logical to explore in more depth bio-inspired approaches to robotics. In particular, where artificial neural systems are implemented using techniques inspired by a greater understanding of how real neurons work. The key underpinning concept of my thesis was that of the cortical self-organising feature map which is how biological brains manage to represent complex information from their environment.  Experimental and computational modelling work, in particular with visual systems, has shown that neurons in the cortex naturally form 2D maps as a representation of many-dimensional input information from the environment. These maps are called ‘feature maps’ as collections of neurons are specialised to detect certain features in an input signal.

As part of this work I developed an autonomous method of regulation of the learning process in the feature map and this was published in Neural Networks - see my Publications page for details.